Order and Chaos in Nature

Course notes
Mahendra K. Verma
August 23, 2010

Contents
1 A bit of history of chaos
1.1 Newton [1642-1727] . . . . . . . . . . . . . . . . . . .
1.2 Laplace [1749-1827]- Determinism . . . . . . . . . . .
1.3 Poincare [1854-1912 ]-Chaos in Three-Body Problem
1.4 Fluid Motion- Weather Prediction [1950] . . . . . . .
1.5 Lorenz - Reincarnation of Chaos . . . . . . . . . . .
1.6 Robert May - Chaos in Population Dynamics . . . .
1.7 Universality of chaos and later developments . . . .
1.8 Deterministic Chaos- Main ingradients . . . . . . . .
1.9 Current Problems of Interest . . . . . . . . . . . . .
1.10 A word on Quantum Mechanics . . . . . . . . . . . .
1.11 Pictures (source: wikipedia) . . . . . . . . . . . . . .
1.12 References . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

4
4
4
5
5
5
6
6
6
6
6
7
7

2 Dynamical System
2.1 Dynamical System . . . . . . . . . . . . . . . .
2.2 Continuous state variables and continuous time
2.3 Continuous state variables and discrete time . .
2.4 Discrete state variables and discrete time . . .
2.5 Discrete state variables and continuous time . .
2.6 Nonlinear systems . . . . . . . . . . . . . . . .
2.7 State Space . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

8
8
8
10
10
10
11
11

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

3 One Dimensional Systems

12
3.1 Fixed points and local properties . . . . . . . . . . . . . . . . . . 12
3.2 Global properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4 Two-dimensional Linear Systems
4.1 Fixed Points and Linear Analysis . . . . .
4.2 Flows in the Linear Systems . . . . . . . .
4.2.1 Real Eigenvalues 1 != 2 . . . . .
4.2.2 Complex Eigenvalues 1,2 = i
4.2.3 Repeated eigenvalues . . . . . . . .
4.3 Damped linear oscillator . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

15
15
16
16
23
26
28

5 Conjugacy of the Dynamical Systems

32
5.1 Linear systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

6 2D Systems: Nonlinear Analysis

6.1 Global Picture . . . . . . . . . . . . . . . .
6.1.1 Example 1: Pendulum . . . . . . . .
6.1.2 Example 2 . . . . . . . . . . . . . . .
6.2 Invariant Manifolds . . . . . . . . . . . . . .
6.3 Stability . . . . . . . . . . . . . . . . . . . .
6.4 No Intersection Theorem and Invariant Sets
6.5 Linear vs. Nonlinear . . . . . . . . . . . . .
6.5.1 Examples . . . . . . . . . . . . . . .
6.6 Dissipation and The Divergence Theorem .
6.7 Poincare-Bendixons Theorem . . . . . . . .
6.7.1 Bendixons Criterion . . . . . . . . .
6.7.2 Poincare-Bendixons Theorem . . . .
6.8 No chaos in 2D systems . . . . . . . . . . .
6.9 Ruling out closed orbits . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

33
33
33
34
36
37
37
37
38
40
41
41
42
43
43

7 Three-Dimensional Systems
7.1 Linear Analysis . . . . . . . . . . . . . . . . . .
7.2 Thm: lin vs. nonlin . . . . . . . . . . . . . . . .
7.3 Example 1: Linear Analysis of Lorenz system .
7.4 Example 2: Two uncoupled oscillator . . . . . .
7.5 Another Example: Forced nonlinear Oscillator
7.6 Poincare Section and Maps . . . . . . . . . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

44
44
45
45
46
47
47

.
.
.
.
.
.
.
.
.
.
.
.
.
.

8 Bifurcation Theory
48
8.1 Bifurcation in 1D Flows . . . . . . . . . . . . . . . . . . . . . . . 48
8.2 Bifurcations in 2D Flows . . . . . . . . . . . . . . . . . . . . . . . 48
9 Homoclinic and Heteroclinic Intersctions
10 One Dimensional Maps and Chaos
10.1 Definitions . . . . . . . . . . . . . .
10.2 Stability of FPs . . . . . . . . . . .
10.3 Periodic orbits and their stability .
10.4 Quadratic Map . . . . . . . . . . .
10.5 Universality of chaos . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

50
.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

11 Bifurcations in Maps

51
51
51
52
52
54
60

12 Characterization of Chaos
12.1 Infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12.2 Fractal Set . . . . . . . . . . . . . . . . . . . . . . . . . .
12.3 Random fractals- . . . . . . . . . . . . . . . . . . . . . . .
12.4 Fractals in nature . . . . . . . . . . . . . . . . . . . . . . .
12.5 Limitations of capacity dimension. Generalized Dimension
12.6 Lyapunov Exponent for maps . . . . . . . . . . . . . . . .
12.7 Lyapunov Exponents for Flows . . . . . . . . . . . . . . .
12.8 Fourier Spectrum . . . . . . . . . . . . . . . . . . . . . . .
13 Intermittency in Chaos

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

65
65
65
65
66
66
67
67
68
69

14 Periodic Orbits In chaos

70
14.1 Sarkovskiis Thm . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
14.2 No of Periodic Orbits and Points . . . . . . . . . . . . . . . . . . 70
15 Quasiperiodic Route to Chaos

72

16 Conclusions

76

A A Brief Overview of Linear Algebra

77

B Nondimensionalization of ODEs

78

C Numerical Solution of ODEs

79
C.1 Numerical Solution of a single ODE- Eulers scheme . . . . . . . 79
C.2 Numerical Solution of a single ODE- Runge Kutta Scheme . . . 81
C.3 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . . . 83

Chapter 1

A bit of history of chaos

1.1

Newton [1642-1727]

Newtons Laws: The equation of motion for a particle of mass m under a force
field F(x, t) is given by
d2 x
m 2 = F(x, t)
dt

Given initial condition x(0) and x(0),

we can determine x(t) in principle.
Using Newtons laws we can understand dynamics of many complex dynamical systems, and predict their future quantitively. For example, the equation
of a simple oscillator is
m
x = kx
whose solution is

!
!
x(t) = A cos ( k/mt) + B sin ( k/mt),

with A and B to be determined using initial condition. The solution is simple

oscillation.
Planetary motion (2 body problem)
r = (/r2 )r,

the solution is elliptical orbit for the planets. In fact the astronomical data
matched quite well with the predictions. Newtons laws could explain dynamics
of large number of systems, e.g., motion of moon, tides, motion of planets, etc.

1.2

Laplace [1749-1827]- Determinism

Newtons law was so successful that the scientists thought that the world is
deterministic. In words of Laplace
"We may regard the present state of the universe as the effect of its past
and the cause of its future. An intellect which at any given moment knew all
of the forces that animate nature and the mutual positions of the beings that
compose it, if this intellect were vast enough to submit the data to analysis,
could condense into a single formula the movement of the greatest bodies of the
universe and that of the lightest atom; for such an intellect nothing could be
uncertain and the future just like the past would be present before its eyes."
4

1.3

Poincare [1854-1912 ]-Chaos in Three-Body

Problem

One of the first glitch to the dynamics came from three-body problem. The
question posed was whether the planetary motion is stable or not. It was first
tackled by Poincare towards the end of ninteenth century. He showed that we
cannot write the trajectory of a particle using simple function. In fact, the
motion of a planet could become random or disorderly (unlike ellipse). This
motion was called chaotic motion later. In Poincares words itself.
If we knew exactly the laws of nature and the situation of the universe at
the initial moment, we could predict exactly the situation of that same universe
at a succeeding moment. but even if it were the case that the natural laws had
no longer any secret for us, we could still only know the initial situation approximately. If that enabled us to predict the succeeding situation with the same
approximation, that is all we require, and we should say that the phenomenon
had been predicted, that it is governed by laws. But it is not always so; it may
happen that small differences in the initial conditions produce very great ones
in the final phenomena. A small error in the former will produce an enormous
error in the latter. Prediction becomes impossible, and we have the fortuitous
phenomenon. - in a 1903 essay "Science and Method.
Clearly determinism does not hold in nature in the classical sense.

1.4

Fluid Motion- Weather Prediction [1950]

Motion of fluid parcel is given by

dv
= p + 2 u.
dt
where , u, and p are the density, velocity, and pressure of the fluid, and is the
kinetic viscosity of the fluid. The above equation is Newtons equation for fluids.
There are some more equations for the pressure and density. These complex set
of equations are typically solved using computers. The first computer solution
was attempted by a group consisting of great mathematician named Von Neumann. Von Neumann thought that using computer program we could predict
weather of next year, and possibly plan out vacation accordingly. However his
hope was quickly dashed by Lorenz in 1963.

1.5

Lorenz - Reincarnation of Chaos

In 1961, Edward Lorentz discovered the butterfly effect while trying to forecast
the weather. He was essentially solving the convection equation. After one run,
he started another run whose initial condition was a truncated one. When he
looked over the printout, he found an entirely new set of results. The results
was expected to be same as before.
Lorenz believed his result, and argued that the system is sensitive to the
initial condition. This accidental discovery generated a new wave in science
after a while. Note that the equations used by Lorenz do not conserve energy
unlike three-body problem. These two kinds of systems are called dissipative
and conservative systems, and both of them show chaos.
5

1.6

Robert May - Chaos in Population Dynamics

In 1976, May was studying population dynamics using simple equation

Pn+1

aPn (1 Pn )

where Pn is the population on the nth year. May observed that the time series
of Pn shows constant, periodic, and chaotic solution.

1.7

Universality of chaos and later developments

In 1979, Feigenbaum showed that the behaviour of Mays model for population
dynamics is shared by a class of systems. Later scientists discovered that these
features are also seen in many experiments. After this discovery, scientists
started taking chaos very seriously. Some of the pioneering experiments were
done by Gollub, Libchaber, Swinney, and Moon.

1.8

Deterministic Chaos- Main ingradients

Nonlinearity: Response not proportional to input forcing (somewhat more

rigourous definition a bit later
Sensitivity to initial conditions.
Deterministic systems too show randomness (deterministic chaos). Even
though noisy systems too show many interesting stochastic or chaotic behaviour, we will focus on deterministic chaos in these notes.

1.9

Current Problems of Interest

A few degrees of system (3 to 6) to Many degrees of systems

Chaotic dynamics: Temporal variation of total population, but systems
with many degrees of freedom show complex behaviour including spatiotemporal phenomena.. Complex Spatiotemporal behaviour is seen in
convection and fluid flows. Turbulent behaviour is observed for even higher
forcing.

1.10

A word on Quantum Mechanics

In QM the system is described by wavefunction. The evolution of the wavefunction is deterministic. However in QM one cannot measure both position and
velocity precisely. There is uncertainty involved all the time
xp h.
Hence, even in QM the world is not deterministic as envisaged by classical
physicists. In the present course we will not discuss QM.
Quantum chaos is study of classical systems that shows chaos.
6

1.11

Pictures (source: wikipedia)

Newton, Laplace, Poincare, Lorenz

1.12

References

H. Strogatz, Nonlinear Dynamics and Chaos, Levant Books (in India).

R. C. Hilborn, Chaos and Nonlinear Dynamics, Oxford Univ. Press.

Chapter 2

Dynamical System
2.1

Dynamical System

A dynamical system is specified by a set of variables called state variables and

evolution rules. The state variables and the time in the evolution rules could be
discrete or continuous. Also the evolution rules could be either deterministic or
stochastic. Given initial condition, the system evolves as
x(0) x(t).
The objectives of the dynamical systems studies are to devise ways to characterize the evolution. We illustrate various types of dynamical systems in later
sections using some examples.
The evolution rules for dyanamical systems are quite precise. Contrast this
with psychological laws where the rules are not precise. In the present course
we will focus on dynamical systems whose evolution is deterministic.

2.2

Continuous state variables and continuous time

The most generic way to characterize such systems is through differential equations. Some of the examples are
1. One dimensional Simple Oscillator: The evolution is given by
m
x = kx,
We can reduce the above equation to two first-order ODE. The ODEs are
x
p

= p/m,
= kx.

The state variables are x and p.

2. LRC Circuit: The equation for a LRC circuit in series is given by
L

dI
Q
+ RI +
= Vapplied .
dt
C
8

The above equation can be reduced to

Q =
LI

I,

Vapplied RI

Q
.
C

The state variables are Q and I.

3. Population Dynamics: One of the simplest model for the volution of
populution P over time is given by
P = P P 2 ,
where is a costant.
A general dynamical system is given by |x(t)& = (x1 , x2 , ....xn )T . Its evolution
is given by
d
|x(t)& = |f (|x(t)&, t)&
dt
where f is a continuous and differentiable function. In terms of components the
equations are
x1

f1 (x1 , x2 , ..., xn , t),

x2 =
.

f2 (x1 , x2 , ..., xn , t)
.

xn

fn (x1 , x2 , ..., xn , t),

where fi are continuous and differentiable functions. When the functions fi are
independent of time, the system is called autonomous system. However, when
fi are explicit function of time, the system is called nonautonomous. The three
examples given above are autonomous sysetms.
A nonautonomous system can be converted to an autonomous oney by renaming t = xn+1 and
x n+1 = 1.
An example of nonautonomous system is
x
p

= p
= x + F (t).

The above system can be converted to an autonomous system using the following
procedure.
x =

p =
t =

x + F (t)
1.

In the above examples, the system variables evolve with time, and the evolution is described using ordinary differential equation. There are however many
situations when the system variables are fields in which case the evolution is described using partial differential equation. We illustrate these kinds of systems
using examples.
9

1. Diffusion Equation

T
= 2 T.
t
Here the state variable is field T (x). We can also describe T (x) in Fourier
space using Fourier coefficients. Since there are infinite number of Fourier
modes, the above system is an infinite-dimensional. In many situations,
finite number of modes are sufficient to describe the system, and we can
apply the tools of nonlinear dynamics to such set of equations. Such
systems are called low-dimensional models.

2. Navier-Stokes Equation
u
+ (u )u = p + 2 u.
t
Here the state variables are u(x) and p(x).

2.3

Continuous state variables and discrete time

Many systems are described by discrete time. For example, hourly flow Qn
through a pipe could be described by
Qn+1 = f (Qn ).
where f is a continuous and single-valued function, and n is the index for hour.
Another example is evolution of the normalized population Pn is the population
at nth year then
Pn+1 = aPn (1 Pn ).

Here the population is normalized with respect to the maximum population to

make Pn as a continuous variable. Physically the first term represents growth,
while the second term represents saturation.
The above equations are called difference equations.
Note that if the time gap between two observations become very small, then
the description will be closer to continuous time case.

2.4

Discrete state variables and discrete time

For some ynamical systems the system variables are discrete, and they evolve
in discrete time. A popular example is game of life where each site has a living
cell or a dead cell. The cell at a given site can change from live to dead or
vise versa depending on certain rules. For example, a dead cell becomes alive
if number of live neighbours are between 3 to 5 (neither under-populated or
over-populated). These class of dynamical systems show very rich patterns and
behaviour. Unfortunately we will not cover them in this course.

2.5

Discrete state variables and continuous time

The values of system variables of logic gates are discrete (0 or 1). However they
depend on the external input that can exceed the threshold value in continuous
time. Again, these class of systems are beyond the scope of the course.
10

In the present course we will focus on ordinary differential equations and

difference equations that deal with continuous state variables but continuous
and discrtet time respectively.

2.6

Nonlinear systems

2.7

State Space
x3
x(t)
x(0)
x2

x3

A state space or phase space is an space whose axis are the state variables
of a dynamical variables. The systems evolution can be uniquely determined
from an initial condition in the state space.

11

Chapter 3

One Dimensional Systems

In this chapter we will consider autonomous systems with one variable. The
evolution of this system will be described by a single first order differential
equation (ODE). In this chapter we will study the dynamics of such systems.

3.1

Fixed points and local properties

The evolution equation of an autonomous one-dimensional dynamical system

(DS) is given by
x = f (x).
The points x at which f (x ) = 0 are called fixed points (FP); at these points
x = 0.
Hence, if at t = 0 the system at x , then the system will remain at x at all
time in future. Noet that a system can have any number of fixed points (0, 1,
2, ...).
Now let us explore the behaviour of the DS near the fixed points:

x f " (x )(x x ),

whose solution is

x(t) x = (x(0) x ) exp (f " (x )t).

Clearly, if f " (x ) < 0, the system will approach x . This kind of fixed point is
called a node. If f " (x ) > 0, the system will go away from x , and the fixed
point is called a repeller. These two cases are shown in the first two figures of
the following diagram:
f(x)

f(x)

f(x)

x
Node

f(x)

x
Repeller

Saddle I

Note that the motion is along the line (along x axis).

12

x
Saddle II

On rare ocassions f (x ) = f " (x ) = 0. In these cases, the evolution near the

fixed point will be determined by the second derivative of f , i.e.,
x

f "" (x )
(x x )2 .
2

If f "" (x ) > 0 (third figure), then x > 0 for both sides of x . If the system is
to the left of x , it will tend towards x . On the other hand if the system is
to the right of x , then it will go further away from x . This point is called
saddle point of Type I. The reverse happens for f "" (x ) < 0, and the fixed point
is called saddle point of Type II.
Examples
1. x = 2x. The fixed point is x = 0. It is a repeller because f " (0) = 2
(positive).
2. x = x+1. The fixed point is x = 1. It is node because f " (1) = 1
(negative).
3. x = (x 2)2 . The fixed point is x = 2. This point is a saddle point of
Type I because f "" (2) = 2 (positive).
The above analysis provides us information about the behaviour near the fixed
points. So they are called local behaviour of the system.

3.2

Global properties

To understand the system completely, we need to understand the global dynamics as well. Fortunately, the global behaviour of 1D systems rather simple, and
it can be easily deduced from the continuous and single valued nature of the
function f . We illustrate the global behaviour using several examples.
Examples:
1. x = x(x 1). FPs are at x = 0 (node) & x = 1 (repeller). Using the single
valued nature of f (x) we can complete the state space diagram which is
shown in Fig.
2. x = x(x 1)(x 2). The FPs are at x = 0, 1, 2. From the information
on the slopes, we deduce that x = 0 and 2 are rellers and x = 1 is node.
These information and continuity of f (x) helps us complete the full state
space picture., which is shown in Fig.
3. x = ax(1 x/k) where a and k are positive constants. The fixed points
are at x = 0 and k. Since f " (0) = a > 0, x = 0 is a repeller. For the other
FP , f " (k) = a < 0, so x = k is a node. Using this information we sketch
the state space, which is shown in Fig....
The above examples show how to make the state space plots for 1D systems.
Using the continuity and single-valued nature of the function f (x) we can
easily deduce the following global properties for a 1D dynamical system.
1. Two neighbouring FPs cannot be nodes or repeller.
2. Two repellers must have a node between them.
13

3. If the trajectories of a system are bounded, then the outermost FP along

the x axis must be (i) node OR (ii) saddle I onthe left and saddle II on
the right.
Note that the above properties are independent of the exact form of f . For
example, systems f (x) = x(x 1) and f (x) = x(x 2) have similar behaviour
even though the forms are different. These properties are called topological
properties of the system.

Exercise
1. For the following systems, obtain the fixed points, and determine their
stability. Sketch state-space trajectories and x t plot
(a) x = ax for a < 0; a > 0.
(b) x = 2x(1 x)
(c) x = x x3

(d) x = x2

(e) x = sin x
(f) x = x cos x

14

Chapter 4

Two-dimensional Linear
Systems
4.1

Fixed Points and Linear Analysis

A general two-dimensional autonomous dynamical system is given by

X 1
X 2

= f1 (X1 , X2 )
= f2 (X1 , X2 ).

The fixed points of the system are the ones where

f1 (X1 , X2 )

= 0

f2 (X1 , X2 )

= 0.

The solution of the above equations yield fixed points which could be one or
more.
Let us analyze the systems behaviour near the fixed point.We expand the
functions f1,2 near the fixed point (X1 , X2 ). Using X1 X1 = x1 and X2 X2 =
x2 , the equation near the FP will be
x 1

x 2

f1 (X1 , X2 )
f1 (X1 , X2 )
|(X1 ,X2 ) x1 +
|(X1 ,,X2 ) x2
X1
X2
f2 (X1 , X2 )
f2 (X1 , X2 )
|(X1 ,,X2 ) x1 +
|(X1 ,X2 ) x2
X1
X2

The four partial derivatives are denoted by a, b, c, d. The above equations can
be written in terms of matrix:
"
# "
#"
#
x 1
a b
x1
=
(4.1)
x 2
x2
c d
Let us look at a trivial system:
# "
#"
#
"
a 0
x1
x 1
=
x 2
x2
0 d
15

whose solution is immediate:

x1 (t) =

x1 (0) exp (at)

x2 (t) =

x2 (0) exp (dt).

When b and/or c nonzero, the equations get coupled. These equations however
can be easily solved using matrix method described below.

4.2

Flows in the Linear Systems

In this section we will solve the equation (4.1) using similarity transformation.
For the following discussion we will use bra-ket notation in which bra (x|
stands a row vector, while ket |x& stands for a column vector. In this notation,
Eq. (4.1) is written as
= A|x&,
|x&
"
#
a b
where A =
is the 2x2 matrix, and |x& = (x1 , x2 )T is a column vector.
c d
The basic strategy to solve the above problem is to diagonalize the matrix A,
solve the equation in the transformed basis, and then come back to the original
basis.
As we will see
" below,#the solution depends crucially on the eigenvalues 1 , 2
a b
of matrix A =
. The eigenvalues of the matrix are
c d
1,2 =

!
1
(T r T r2 4)
2

with T r = a + d is the trace, and = ad bc is the determinant of the matrix.

These eigenvalues can be classified in four category
1. Real ones with 1 != 2 (when T r2 > 4).
2. Complex ones {1,2} = (when T r2 < 4).
3. 1 = 2 with b = c = 0 (here T r2 = 4).
4. 1 = 2 with b != 0, c = 0 (here T r2 = 4).
We will solve Eq. (4.1) for the four above cases separately.

4.2.1

Real Eigenvalues 1 != 2

The eigenvectors corresponding to the eigenvalues 1,2 are |v1 & = (1, b/(a
1 )T and |v2 & = (1, b/(a 2 )T respectively. Since 1 != 1 , |v1 & and |v2 &
are linearly independent. Using these eigenvectors we construct a nonsingular
matrix"S whose
columns"are #
|v1 & and |v2 &. We denote the unit vectors by
#
1
0
|e1 & =
and |e2 & =
. Clearly S|e1 & = |v1 & and S|e2 & = |v2 &.
0
1
The matrix S 1 AS
S 1 AS|e1 & = S 1 A|v1 & = 1 S 1 |v1 & = 1 |e1 &.
16

Similarly

S 1 AS|e2 & = 2 |e2 &.

Therefore S 1 AS is a diagonal matrix whose diagonal elements are the eigenvalues of A, i.e.,
"
#
1 0
S 1 AS = D =
.
0 2

This procedure is called diagonalization of matrix. Note that the whole proof
hinges on the existance of two linearly indendent eigenvectors.
In the following we will use the above theorem to solve the DE. Inversion of
the above equation yields A = SDS 1 and
= A|x& = SDS 1 |x&.
|x&

Using S 1 |x& = |y&, we obtain a much simpler looking equation

= D|y&,
|y&
whose solution is
|y(t)& = y1 (0) exp (1 t)|e1 & + y2 (0) exp (2 t)|e2 &.

Using |x& = S|y&, we obtain

|x(t)& = y1 (0) exp (1 t)|v1 & + y2 (0) exp (2 t)|v2 &.

We can derive the above solution in another way. The solution of Eq. (4.1)
in the matrix form is
|x(t)& = exp (At)|x(0)&.

Since

exp (At) = S exp (Dt)S 1 ,

we obtain
exp (At)|x(0)&

=
=

S exp (Dt)|y(t)&
S exp (Dt)[y(0)|e1 & + y2 (0)|e1 &]

y1 (0) exp (1 t)|v1 & + y2 (0) exp (2 t)|v2 &,

which is same as the above result.

Graphically in y1 -y2 coordinates the solution is

Elimination of t yields

y1

y1 (0) exp (1 t)

y2

y2 (0) exp (2 t).

y2 (t) = C[y1 ]2 /1 ,

where C is a constant. We can plot these the phase space trajectories in y1 -y2
plane very easily. When 1,2 are of both positive (1,2 here), then we obtain
curves of the type I (see Fig. 4.1). The fixed point is called a repeller and the
phase space curves are y2 = Cy12 . When the eigenvalues are both negative, the
the fixed point is a node; the flow is shown in Fig. 4.2 (-1,-2 here) with phase
17

y2(t)

3
3

0
y1(t)

Figure 4.1: A phase space plot of a system y1 = y1 , y2 = 2y2 . The fixed point
is a repeller.
space curves again as y2 = Cy12 . In Fig. 3, 1 is positive, while 2 is negative,
with the fixed point termed as saddle ( =1,-1 respectively). The flow diagram
for a saddle is shown in Fig. 4.3 with the phase space curves as y2 y12 = C.
The y1 and y2 axes are the eigen directions in y1 -y2 plane. If the initial condition lies on either of the axes, then the system will remain on that axis forever.
Hence these are the invariant directions.
In x1 -x2 plane the above mentioned state space plots are similar to those in
y1 -y2 plane except that the eigen directions are rotated. This is illustrated by
the following example:
Example 1:
"
#
1 0
A=
1 2

T r = 3, = 2. The eigenvalues
1,2 =

98
= 1, 2.
2

The eigenvectors corresponding to these values are

"
#
"
#
1
0
; |v2 & =
|v1 & =
1
1
respectively.Hence
S=

"

1
1

18

0
1

y2(t)

3
3

0
y1(t)

Figure 4.2: A phase space plot of a system y1 = y1 , y2 = 2y2 . The fixed

point is a node.

y2(t)

3
3

0
y1(t)

Figure 4.3: A phase space plot of a system y1 = y1 , y2 = 2y2 . The fixed point
is a saddle.

19

y (t)

3
3

0
y (t)

Figure 4.4: A phase space plot of a system x1 = x1 , x2 = x1 2x2 . The fixed

point is a node.

y2(t)

3
3

0
y1(t)

Figure 4.5: A phase space plot of a system x1 = 4x1 + 2x2 , x2 = 2x1 + 4x2 . The
fixed point is a repeller.

20

y2(t)

3
3

0
y1(t)

Figure 4.6: A phase space plot of a system y1 = y2 , y2 = y1 . The fixed point is

a saddle.
It is easy to verify that
S 1 AS =

"

1 0
0 2

In the eigen basis the solution is

|y(t)& = y1 (0) exp (t)|e1 & + y2 (0) exp (2t)|e2 &
which is shown in Fig. 4.2. The constants y1,2 (0) can be obtained from the
initial condition.
In the x1 x2 basis the solution is
|x(t)& = y1 (0) exp (t)|v1 & + y2 (0) exp (2t)|v2 &.
The flow in x1 x2 is show in Fig. 4.4. If
direction (either on |v1 & or |v2 &), then the
axis forever.
Example 2:
"
4
A=
2

the initial condition lies on the eigen

system will contiue to remain on the
2
4

The eigenvalues of the matrix are 2 and 6, and the corresponding eigen vectors
are (1, 1)T and (1, 1)T . The phase space picture in x1 -x2 basis is shown in
Fig. 4.5.
Example 3: Motion in a potential U (x) = x2 /2
Equation of motion
dU
= x.
x
=
dx
Therefore,
x 1
x 2

=
=
21

x2 ,
x1 .

Therefore the matrix A is

A=

"

0
1

1
0

whose
" 1 and
# -1. The corresponding eigenvectors are |v1 & =
"
# eigenvalues are
1
1
respectively. These are the diagonals as shown in
and |v2 & =
1
1
Fig. 4.6. The eigenvalue is +1 long the eigen direction |v1 &, so the system will
move away from the origin as exp(t). However the system moves towards the
the origin along |v2 & as exp(t) since the eigenvalue is -1 along this direction.
In the x1 x2 basis the solution is
|x(t)& = y1 (0) exp (t)|v1 & + y2 (0) exp (t)|v2 &.

After some manipulation we can show that

x1

y1 + y2 ,

x2

y1 y2 .

One can easily show that the DEs are decoupled in y1,2 variables.
We can eliminate t and find the equations of the curves
y1 y2 = C,
which are hyperbola. In terms of x1 x2 , the equations are
x21 x22 = C " .

The flow are shown in Fig. 4.6.

The equation of the curves could also be obtained using
x
dp
= ,
dx
p
or

p2
x2

= const = E.
2
2
The above equation could also be obtained from the conservation of energy.
The curves represent physical trajectories. See figure below. Note that E < 0
and E > 0 have very different bahviour. The curve E = 0 are the eigenvectors.
Interpret physically. Show that when E = 0,the system take infinite time to
reach to the top.

D
A
E=5
R

E = 5

E=0

E = 5

E=0
T

U
P

E=5
B

0
x

Our system is in region A at point P shown in the figure. See the directions
22

of the arrow. The fixed point is (0, 0). Technically this type of fixed point is
called a saddle. All the unstable fixed points of mechnaical systems have this
behaviour.

4.2.2

Complex Eigenvalues 1,2 = i

Let us solve the oscillator whose equation is

x
= x
or
x 1
x 2
Clearly the matrix A is
A=

=
=
"

x2
x1 .

0 1
1 0

whose eigenvalues are i. The corresponding eigenvectors are

" # "
#
1
0
;
.
i
i
It is quite inconvenient to work with complex vectors. In the following discussion
we will devise scheme to use real vectors to solve DEs whose eigenvalues are
complex.
Along the first eigenvector the solution is
" #
1
|x& = exp (it)
i
"
#
"
#
cos t
sin t
=
+i
sin t
cos t
= |xre & + i|xim &

= A|x& easily yields

The substitution of the above in |x&
& =
|xre
& =
|xim

A|xre &,

A|xim &,

Hence, |xre & and |xim & are both solution of the original DE. Since |xre & and
|xim & are linearly independent solutions of the DE, we can write the general
solution x(t) as a linear combination of |xre & and |xim &:
|x&

= c1 |xre & + c2 |xim &

"
#
"
#
cos t
sin t
+ x 1 (0)
.
= x1 (0)
sin t
cos t

Hence
x1 (t)
x2 (t)

=
=

x1 (0) cos t + x2 (0) sin t

x1 (0) sin t + x2 (0) cos t.
23

y2(t)

3
3

0
y (t)

Figure 4.7: A phase space plot of a system x1 = x2 , x2 = x1 . The fixed point

is a node.
Note that x2 (t) = x 1 is the velocity of particle. Both the position and velocity
are periodic, as is expected for oscillatior. Clearly
x21 + x22 = c21 + c22 = C.
These are the equations of concentresic circles. See the following figure and
2

E=1

1.5
1

0.5
0

A
C

0.5
E = 0.5

1.5
2
1

0
x

Note that we could have had derived the equations for the curve using the
energy conservation.
"
#
1
Recall that the independent solutions for real eigenvalues were exp (1 t)
0
"
#
0
and exp (2 t)
, which corresponds to the motion along the eigen direction.
1
For the above case with imaginary eigenvalues, the eigenvectors are"not along #a
cos t
straight line in the real space. Here the system essentially moves as

sin t
"
#
sin t
and
which corresponds to the clockwise and anticlockwise circular
cos t
motion respectively.
In the above discussion we have shown how to use real vectors to solve the
DEs whose eigen values are pure imaginary. In the following we will extend the
procedure to complex eigenvalues. Suppose the eigenvalues are i, and the
eigenvector corresponding to + i is |v& = |v1 & + i|v2 & where |v1 & and |v2 & are

24

real vectors. Using A|v& = ( + i)|v& we obtain

A|v1 & =
A|v2 & =
Let

|v1 & |v2 &

|v1 & + |v2 &.

S = (|v1 & |v2 &),

then

S 1 AS|e1 & =
S

AS|e2 & =

Therefore
S

AS =

"

|e1 & |e2 &

|e1 & + |e2 &.

= B,

and the equations in terms of the transformed variables are

= B|y&.
|y&

(4.2)

Note that |v & is the other independent eigen vector. we can also diagonalize A
). However we wish to avoid the complex vectors.
by using (
vv
In the following discussion we will"solve
# Eq. 4.2. In the new basis the eigen1
vector corresponding to ( + i) is
. Along this direction the evolution
i
is
" #
1
|y(t)& = exp (t) exp (it)
i
"
#
"
#
cos (t)
sin (t)
= exp (t)
+ i exp (t)
sin (t)
cos (t)
= |yre & + i|yim &.

We can immediately derive that

&
|yre
&
|yim

= B|yre &,

= B|yim &,

Hence, |yre & and |yim & are both independent solution of Eq. 4.2, and the general
solution |y(t)& as a linear combination of |yre & and |yim &:
|y(t)&

= c1 |yre (t)& + c2 |yim (t)&

"
#
"
#
cos (t)
sin (t)
= x1 (0) exp (t)
+ x2 (0) exp (t)
.
sin (t)
cos (t)

The equations of the trajectories are

$ 2
%
y1 + y22 exp (2t) = c21 + c22 = C,

which are the equations of spirals as shown in Fig. 4.8. When < 0, the system
moves towards the origin (fixed point), and the fixed point is called spiral node.
However when > 0, the system moves outward, and the fixed point is called
a spiral repeller.
25

y (t)

3
3

0
y1(t)

Figure 4.8: A phase space plot of a system x1 = 0.5x1 +x2 , x2 = x1 +0.5x2.

The fixed point is a spiral node.

4.2.3

Repeated eigenvalues

Consider matrix
A=

"

a
c

b
d

The condition for the eigenvalues to be equal is T r2 = 4, which implies that

(a d)2 = 4bc.
If the eigenvector is (u1 u2 )T , then we obtain

bcu1 + bu2 = 0.

(4.3)

(4.4)

The above two conditions can be satisfied in the following cases:

1. a = d and b = c = 0: Here any vector is an eigenvector and we can choose
any two independent ones for writing the general solution of DE.
2. a = d, c = 0 but b != 0: Here only one eigenvector (1, 0)T exists.
3. a = d, b = 0 but c != 0: This is same as case 2.
4. a != d: Here both b and c are nonzero, hence there will only be one
eigenvector.
The above four cases essentially fall into two cases: (a) having two indendent
eigenvectors; (b) haveing only one eigenvector.In the following discussions we
= A|x& under the above two cases.
will study the solution of |x&

26

y (t)

3
3

0
y (t)

Figure 4.9: A phase space plot of a system x1 = 2x1 , x2 = 2x2 . The fixed point
is a repeller.
With two independent eigenvectors: 1 = 2 , b = c = 0
According to the above discussion, the matrix A will be of the form
"
#
a 0
A=
= aI.
0 a
The solution is trivial.
x(t) = x1 (0) exp (at)E1 + x2 (0) exp (at)E2 .
The equation of the curves are
x2 (t) = Cx1 (t)
as shown in Fig. 4.9. Note that anytwo linearly independent eigenvectors are
eigenvectos of the matrix.
With only eigenvector: 1 = 2 , b != 0, c = 0

Using CayleyHamilton theorem theorem we have

(A I)2 |w& = 0

(4.5)

for all |w&. Suppose the one eigenvector is |v& then

(A I)|v& = 0.
We can expand any vector in a plane using |v& and another linearly independent
vector, say |e&. Hence
|w& = |v& + |e&.
27

y2(t)

3
3

0
y (t)

Figure 4.10: A phase space plot of a system y1 = 2y1 + y2 , y2 = 2y2 . The fixed
point is a repeller.
Substitution of the above form of |w& in Eq. (4.5) yields
(A I)|e& = |v&.
If = 0, |e& will also be a eigenvector, which is contrary to our assumption.
Hence != 0. Using |e&/ = |u&, we obtain
A|u& = |v& + |u&.
Using

S = [|v& |u&]

we obtain

S 1 AS =

"

= B,

which is the Jordon-Cannonical form for the 2x2 matrix with repeated eigenvalues. In the new basis

whose solution is
|x(t)& =

y 1

y1 + y2

y 2

y2

"

exp (t)
0

t exp (t)
exp (t)

We summarize various fixed points using Tr- plots.

4.3

Damped linear oscillator

The equation of a damped linear oscillator is

x
+ 2 x + x = 0

28

y2(t)

3
3

0
y1(t)

Figure 4.11: A phase space plot of a system y1 = 2y1 + y2 , y2 = 2y2 . The

fixed point is a node.

y2(t)

3
3

0
y1(t)

Figure 4.12: A phase space plot of a system x1 = 3x1 + x2 , x2 = x1 + x2 . The

fixed point is a repeller.

29

y (t)

3
3

0
y1(t)

Figure 4.13: A phase space plot of a system x1 = x2 , x2 = x1 + 2x2 with

= 0.1. The fixed point is a spiral node.
which reduced to
x1

x2

x2

x1 2x2 .

The eigenvalues for the system are

1,2 =
Clearly

!
2 1.

For < 1, the eigenvalues are complex.

For > 1, the eigenvalues are real and negative
For = 1, the eigenvalues are real and repeated.
The state space plots for these three cases are shown in the following three
figures. Find out the eigenvectors for these cases.

30

y2(t)

3
3

0
y1(t)

Figure 4.14: A phase space plot of a system x1 = x2 , x2 = x1 + 2x2 with

= 2. The fixed point is a node.

y (t)

3
3

0
y1(t)

Figure 4.15: A phase space plot of a system x1 = x2 , x2 = x1 + 2x2 with

= 1. The fixed point is a node.

31

Chapter 5

Conjugacy of the Dynamical

Systems
Suppose the two linear systems x = Ax and y = By have flows A and B
respectively. These two systems are (topologically) conjugate if there exists a
homeomorphism h : R2 R2 that satisfies
B (t, h(X0 )) = h(A (t, X0 )).
The homeomorphism h is called a conjugacy. Thus a conjugacy takes the solution curves of x = Ax to y = By.
Example:
Two systems x = 1 x and y = 2 y.
We have flows
j (t, x0 ) = x0 exp (j t)
for j = 1, 2. If 1,2 are nonzero and have the same sign, then
&
x2 /1 for x 0
h(x) =
|x|2 /1 for x<0
Hence the two systems are conjugate to each other. This works only when 1,2
have the same sign.

5.1

Linear systems

Def: A matrix A is hyperbolic if none of its eigenvalues has real part 0. We also
say that the system x = Ax is hyperbolic.
Thm: Suppose that the 2 2 matrices A1 and A2 are hyperbolic. Then the
linear systems x = Ax are conjugate if and only if each matrix has the same
number of eigenvalues with negative real parts.
Without proof.

32

Chapter 6

2D Systems: Nonlinear
Analysis
The nonlinear system is given by
x = f (x),
where f (x) is a nonlinear function. In 2D we write
x =
y =

f (x, y)
g(x, y)

with f and g as nonlinear functions of x and y. Let us denote the domain by

D. D R2 .

6.1
6.1.1

Global Picture
Example 1: Pendulum

The nondimensionalized equation of a pendulum is

= sin ,
where is the angle from the stable equilibrium position in the anticlockwise
direction. We can rewrite the above equation as

sin .

The DS has two fixed points: (0, 0) and (, 0) (because the system is periodic
in ).
Linearization near (0, 0) yields

33

y2(t)

4
4

0
y1(t)

Figure 6.1: A phase space plot of a system y1 = y1 , y2 = 2y2 . The fixed point
is a repeller.
Clearly the fixed point is a center. It is consistent with the fact that the (0, 0)
is a stable equilibrium point. Linearization near (0, )yields
=
v =

v
,

which is a saddle. Consistent because (, 0) is an unstable equilibrium point.

The eigen vectors at (, 0) are (1,1) and (1, 1) (evs. 1, -1). Sketch the linear
profile.
Now let us try to get the global picture. The conservation of energy yields
v2
+ (1 cos ) = E
2
!
v = 2(E 1 + cos .

or

The above function can be plotted for various values

of E. For E = 0 we get a
point (0, 0). For E = 1 we get the curves v = 2 cos which passes through
the saddle point. These curves are called separatrix for the reasons given
below. For E < 1, the curves are closed and lie between the two separatrix.
Above E = 1, the curves are open as shown in the figure. The separatrix
separate the two set of curves with differen qualitative behaviour.
Q: How long does it take for the pendulum to reach to the top when E = 1.
Note that the above curves nicely join the lines obtained using linear analysis.
Something similar happens for the following system as well.

6.1.2

Example 2

The equation of motion is

x = x + x3 .
34

y2(t)

3
3

0
y1(t)

Figure 6.2: A phase space plot of a system y1 = y1 , y2 = 2y2 . The fixed point
is a repeller.
The force (p)
is zero at x = 0, 1. Therefore, the fixed point are (0, 0) and
(1, 0) (note x = 0). The potential plot is given in the figure:
1.5
E = 1/2

0.3

0.5

0.2

0.1

0.5

E = 1/4

U(x)

E = 7/64
0

0.1
0.2
1.5

1 0.5 0 0.5
x
(a)

1.5

1.5 1

0.5 0
x

0.5

1.5

(b)

Clearly (0, 0) is a stable FP, and (1, 0) are unstable FP. Near (0, 0), the
equation are
x =
p =

p,
x,

which is the equation of the oscillator. Hence, near (0, 0) the behaviour will be
the same as that near the oscillator. Now near (1, 0), change the variable to
x" = x 1. In terms of (x" , p), the equations are
x
p

=
=

p,
2x" ,

which is the equation for the unstable hill discussed in Problem 3.5.5. Hence,
the phases space around (1, 0) should look like of Problem 3.5.5. The same thing
for (1, 0) also. You can see from the potential plot that (1, 0) are unstable
35

points. Therefore, the phase space plot will look like as shown in the figure.
What are the other phase trajectories doing?
(b) By similar analysis we draw the phase space trajectory for a DS whose
equation is
x
= x x3 .
1.5
1

0.5
0

0.5
1
1.5
1

6.2

0
x

Invariant Manifolds

A Manifold is a subspace of the state space that satisfies continuity and differentiability property. For example, fixed points, x-axis, etc. are manifolds. Among
these, invariant manifolds are special. If the initial condition of a DS starts
from an point on a manifold and stays within it for all time, then the manifold
is invariant. For linear systems, the fixed points and eigen vectors are invariant
manifolds. The system tends to move away from the fixed points on the eigenvectors corresponding to the positive eigenvalues, hence these eigenvectors are
unstable manifold. The eigenvectors with negative eigenvectors are stable
manifold since the system tends to move toward the fixed points if it starts on
these curves.
For nonlinear systems, the fixed points are naturally invariant manifolds.
However the stable and unstable manifolds are not the eigenvectors of the linearized matrix of the fixed points. Yet we can find the stable and unstable
manifolds for the nonlinear systems using the following definitions.
Def: The stable manifold is the set of initial conditions |x0 & such that
|x(t)& |x & as t where |x & is the fixed point. Similarly unstable
manifold is the set of initial conditions |x0 & such that |x(t)& |x & as t
(going backward in time).
For the pendulum and x2 x4 potential (Figs), the fixed points and the
constant energy periodic curves are invariant manifolds. In addition stable
and unstable manifolds for the saddle are also visible. Incidently, the unstable
manifold of one saddle merges with the stable manifold of the other saddle.
Trajectories of these kind that join two different saddles are called heteroclinic
trajectories or saddle connection. For x2 +x4 potential (Fig), the unstable
manifold of the saddle merges with its own stable manifold; such trajectories
are called homoclinic trajectories (orbits).
36

6.3

Stability

A FP is stable equilibrium if for every neighborhood O of x in Rn there is

a neighbourhood O1 of x such that every solution x(t) with x(0) = x0 in O1 is
defined and remains in O for all t > 0. This condition is also called Liapunov
stability.
In addition, if limt x(t) = x , then the FP is called asymptotically
stable (or attracting).
The FPs which are not stable are called unstable FP.
Examples and some important points to remember
Center- stable but not asymptotically stable
Node- asymptotically stable
Saddle, repellers- unstable
For nonlinear systems, the stability is difficult to ascertain.
Liapunov stability does not imply asymptotic stability (e.g., center). When
a system is Liapunov stable but not asymptotic stable, it is called neutrally
stable.
Asymptotic stability does not imply Liapunov stability. Consider a DS
= 1 cos . The asymptotic fixed point is = 0, but f " ( = 0) = 0 and
f "" ( = 0) > 0. So the systems is approaches = 0 from left,. However
for any initial condition > 0, the system increases to = 2 and reaches
the fixed point due to its periodic nature.

6.4

No Intersection Theorem and Invariant Sets

Two distinct state space trajectories cannot intersect in a finite time.

Also, a single trajectory cannot cross itself.
Proof: Given an initial condition, the future of the system is unique. Given
this we can prove the above theorem by a proof of contradiction. It the state
space trajectories interesect in a finite time, then we can take the point of
intersection to be the initial point for the future evolution. Clearly, at the point
of intersection there will be two directions of evolution, which is a contradiction.
Hence No Intersection Theorem.
Note that the trajectories can intersect at infinite time. This point is a
saddle.

6.5

Linear vs. Nonlinear

Hartman-Grobman Theorem: Suppose the ndimensional system x =

F (x) has an equilibrium point at x0 that is hyperbolic. Then the nonlinear
flow is conjugate to the flow of the linearized system in a neighbourhood of x0 .
s
In addition, there exist local stable and unstable manifolds Wloc
(x0 ) and
u
Wloc (x0 ) of the same dimension ns and nu as those of the eigenspaces E s and
E u of the linearized equations, and the manifolds are tangent to E u and E s .

37

6.5.1

Examples

(1) DS
x
y

= x + y2
= y

Locally: saddle
General solution
x(t)
y(t)

1
1
= (x0 + y02 ) exp t y02 exp (2t)
3
3
= y0 exp t

Sketch the plot. The stable manifold is the same as E s . However the unstable
manifold is
x + y 2 /3 = 0.
Locally the same behaviour as the linearized system.
(2) DS
x =
y =

y + x( r2 )
x + y( r2 )

Linear spiral, nonlinearr

=
=

r( r2 )
1.

Same behaviour near the fixed point.

(3) DS
x

= y + ,xr2

= x + ,yr2 .

The linear solution- center.

Nonlinear eqns
r

,r3

1.

which is a spiral. So near the FP the linear behaviour si very different then the
nonlinear behaviour.
(4) DS
x =

x2

The nonlinear soln

x(t)

y(t) =

x0
1 x0 t
y0 exp (t)

The yaxis is the stable manifold.

The linear behaviour is x(t) = const. The nonlinear and linear bahaviour
very different.
The above examples illustrate Hartman-Grobman theorem.
38

y2(t)

3
3

0
y1(t)

Figure 6.3: A phase space plot of Example 1

y2(t)

3
3

0
y1(t)

Figure 6.4: A phase space plot of Example 3; outward spiral

39

y (t)

3
3

0
y (t)

Figure 6.5: A phase space plot of Example 4

6.6

Dissipation and The Divergence Theorem

Let us look at the time evolution of an area. Consider an are given in the figure.
Its area is
A = dxdy = (xB xA )(yD yC ).
By differentiating the above, we obtain
dA
dt

(xB xA )[g(xD , yD ) g(xC , yC )] + [f (xB , yB ) f (xA , yA )](yB yA )

dxdy

f
g
+ dxdy .
y
x

Hence,

1 dA
= div(f, g).
A dt
The above theorem can be easily generalized to any dimenison as
1 dV
= f.
V dt
If the div < 0, the the volume will shrink, and if the div > 0, then the volume
will grow. The flows with div = 0 are area preserving. The systems with div<0
are called dissipative systems, and ones which are area preserving are called
Hamiltonian systems.
Examples:
1. Show that all the mechanical systems which can be described by positiondependent potentials are area-preserving. Demonstrate using SHM as an
example.
2. Show that state-space of frictional oscillator (positive friction) is dissipative.

40

3. Consider
x

sin x(0.1 cos x cos y)

sin y(cos x 0.1 cos y).

Describe the motion.

Note that the div is trace of matrix A for the linear system or linearized
system near the fixed point. Since trace is invariant under similarity transformation:
div(f, g) = T r(A) = 1 + 2 .

6.7

Poincare-Bendixons Theorem

6.7.1

Bendixons Criterion

Suppose that the domain D R2 is simply connected (no holes or separate

parts in the domain) and f and g continuously differentiable in D . The system
can only have periodic solutions if (f, g) = 0 or if it changes sign. If the div
is not identically zero or it does not changes sign in D, then the system has no
closed solution lying entirely in D.
Proof: If we have a closed orbit C in D. The interior of D is G. Then Gauss
law yields
'
'
'
dx
dy
g )dt = 0,
(f, g)d =
(f dy gdx) = (f
dt
dt
G
C
C
which is possible only if (f, g) = 0 everywhere or if it changes sign. If (f, g)
has one sign throughout R, then the above condition will not be satisfied, hence
no closed orbit will be allowed. Note that this is a necessary condition, but not
a sufficient condition.
Examples
1. Van der Pol oscillator

x
+ x (1 x2 )x

with as a constant. The DS can be rewritten as

x =
y =

y
x + (1 x2 )y.

Clearly the divergence is (1 x2 ). For |x| < 1, the div is nonzero and
monotonic, hence this region cannot have a periodic solution in the domain
|x|<1 .
2. A DS
x =
y =

x + y 2
y 3 + x2 .

div = 1 3y 2 , which is clearly negative. Hence DS does not have a

periodic solution.
41

3. SHM: div=0. Periodic solution may exist. In fact, the system has infinite
periodic orbits.
4. For Hamiltonian systems
x

= y

dU
dx
where U is the potential. Clearly (f, g) = 0. But Hamiltonian system
need not have a closed orbit. For example, U = x2 . This example illustrates that Bendixons criterion is a necessary condition for the existance
of closed orbit, but not a sufficient condition.
y

6.7.2

Poincare-Bendixons Theorem

= |f (|x&)& in a region R with |f (|x&)& =

Consider a two-dimensonal system |x&
T
(f, g) being a continuous and differentiable vector field. If R is a closed and
bounded subset of the plane, and if a trajectory C is confined in R, then either
C is a closed orbit;

C approached a closed orbit asymptotically as t ;

C approaches a fixed point asymptotically as t .

If the DS does not have a fixed point then the system is guaranteed to have an
isolated closed orbit called limit cycle.
Example:
x =
y =

y + x( r2 )
x + y( r2 )

with > 0. Bounded system. 0 is a repeller spiral. Since there is no other fixed
point, the system must have a LC.
Example A DS is given by
x =
y =
Hence the div is
div

x(x2 + y 2 2x 3) y
y(x2 + y 2 2x 3) + x.

3(x2 + y 2 ) + y 2 + x2 6x 3 3
(
)
3
33
4 (x )2 + y 2
.
4
16

=
=

The
bracketed term is an equation of a circle with center at (3/4, 0) and radius
33/4. The sign of div is negative for all the points within the circle. So, using
Bendixons criterion, there cannot be a periodic orbit within the circle.
In radial polar coordinate
r

=
=

r(r2 2r cos 3)
1.

The term in the bracket is positive for r > 3, mixed for 1 < r < 3, and negative
for r < 1. Then the annulus 1 < r < 3 should have a limit cycle. It will be an
unstable LC.
42

6.8

No chaos in 2D systems

Chaos- divergence of trajectories in bounded systems.

No Intersection thm. prohibits the diverging trajectories to come back.
Hence no chaos in 2D. The diverging trajectories keeps diverging and moving
away.

6.9

Ruling out closed orbits

Strogatz Sec. 7.2 p199.

Gradient system
Liapunov function
Energy function

43